Nanbeige4.1

Nanbeige LLM Lab, Boss Zhipin · 2026-02-13 · analysed 2026-08-10 · skill v0.3.5

dense text-only open-weights medium frontier-generalist

Design overview

Nanbeige4.1-3B is a post-training recipe applied to the Nanbeige4-3B-Base checkpoint; the report discloses no architecture beyond the 3B parameter count and a 256K context, so the block itself (a dense decoder, per family context) is inherited unchanged. The contribution is a cascaded, capability-balanced training stack. SFT is re-mixed toward code and hard math, run through a three-stage context curriculum (32k → 64k → 256k, with the final stage mixing code 27%, deep-search 26%, STEM 23%, tool-use 13%, general 10%), and improved by scaling the Solution Refinement / CoT Reconstruction loop from the prior Nanbeige4-3B-2511 release. General RL then runs GRPO twice: point-wise RL with a general reward model (8 rollouts per prompt) that suppresses redundancy and formatting errors, then pair-wise RL with a binary win/loss reward and a swap-consistency regularizer to remove position bias. Code RL runs in two stages — correctness via a pass-rate reward, then a gated time-complexity reward active only when all tests pass, with a judge system (multi-language sandbox plus an instruct judge model) comparing against reference optimal complexity. Finally a lightweight agentic RL stage, trained on Wikipedia-graph multi-hop QA and turn-judged long trajectories (open-sourced as ToolMind-Web-QA), delivers stable long-horizon search — up to 600 tool-call turns. The result is a 3B model that leads small models on deep-search benchmarks and beats much larger peers on coding and alignment — at the cost of disclosing no architectural detail at all.

What the report claims is novel

Atlas assessment: A pure post-training contribution on the Nanbeige4-3B base: no architectural novelty is claimed or described (the report has no architecture section at all). The reward-engineering stack — point-wise + pair-wise RL, a gated time-complexity reward, and turn-level vs trajectory-level credit in agentic RL — extends the cascaded-RL line the report cites (Nemotron-Cascade); the gated complexity reward that fires only on fully correct solutions is the most distinctive element. The 600-turn agentic-stability and deep-search claims at 3B scale are strong but not independently verifiable. vs microsoft-phi-4 (the atlas's closest 'small generalist' peer): same small-model-broad-capabilities intent, but Nanbeige4.1 optimizes for agentic deep search and code efficiency rather than instruction-following density; vs weibo-vibethinker-3b: same 3B reasoning scale, but Nanbeige adds long-horizon tool use and alignment stages. The complete absence of architecture disclosure makes cross-atlas block comparison impossible.

Relation to the atlas

Lineage in

Influence out

No descendants recorded yet.

Notable omissions

Closest relatives in the atlas

Spec sheet

Identity

FamilyNanbeige4.1[1]
OrganisationNanbeige LLM Lab, Boss Zhipin[2]
Release date2026-02-13
Report URLhttps://arxiv.org/abs/2602.13367[3]
Licensenot disclosed
Open weightsyes[4]

Scale

Total parameters3B[5]
Active parameters / token3B
Layersnot disclosed
Hidden dimensionnot disclosed
FFN inner dimensionnot disclosed
FFN rationot disclosed
Query headsnot disclosed
KV headsnot disclosed
Head dimensionnot disclosed
Vocabulary sizenot disclosed
Embedding tyingnot disclosed

Core block

Block typedense
Attention variantnot disclosed
Attention layer patternnot disclosed
Depth mixingsequential-residual
Activationnot disclosed
Expert countn/a[6]
Experts per tokenn/a[6]
Shared expertsn/a[6]
Routingn/a[6]
Load balancingn/a[6]
Expert granularityn/a[6]
Methodnot disclosed
RoPE base frequencynot disclosed
Partial RoPEnot disclosed
Typenot disclosed
Placementnot disclosed
QK-normnot disclosed
Attention sinksnot disclosed
Softcappingnot disclosed
Othernot disclosed

Context

Trained context length262,144[7]
Deployed context length262,144
Extension methodnone

Tokenizer

Algorithmnot disclosed
Notesnot disclosed

Training

Training tokensnot disclosed
Data compositionSFT mixture redesigned with a higher proportion of code-related data and higher ratios of challenging mathematics and general-domain problems; final 256k-stage mixture: code 27%, deep-Search 26%, STEM 23%, tool-use 13%, general domains 10%. Deep-search data: multi-hop QA pairs from Wikipedia entity-relation graphs (temporal-aware head-entity selection, conditional random walks) plus synthesized long-horizon trajectories with turn-level judgment (open-sourced as ToolMind-Web-QA). Code data: judge-system-filtered time-optimal solutions; RL data: point-wise (human preference), pair-wise (strong-weak model comparisons with checklist filtering), code (problem + test cases + time-complexity-optimal solution + optimal complexity label), agentic (1,827 synthesized task-oriented environments with tools and verifiers)[8]
Curriculum / stagingpost-training pipeline on the Nanbeige4-3B-Base: extended SFT (three context stages, final 256k) → point-wise RL (GRPO, 8 rollouts per prompt, general reward model) → pair-wise RL (pairwise reward model, swap-consistency regularizer) → two-stage code RL (Stage 1 correctness pass-rate reward; Stage 2 gated time-complexity reward active only at PassRate=1) → lightweight agentic RL; on-policy difficulty filtering in code RL (n=8 rollouts, retain problems with k in [1,5] qualifying rollouts)[9]
Optimizernot disclosed
LR schedulenot disclosed
Batch schedulenot disclosed
Precisionnot disclosed
Parallelismnot disclosed
Hardwarenot disclosed
Disclosed computenot disclosed

Post-training

SFTyes[10]
Preference optimisationGRPO[11]
Reasoning trainingcascaded multi-stage RL: point-wise RL with a general reward model (8 rollouts per prompt) suppressing redundancy/repetition (overlong truncation reduced from 5.27% to 0.38% on LiveCodeBench-v6); pair-wise RL with binary win/loss reward vs reference answers, multi-turn dialogue history concatenated into the reward model input; two-stage code RL with correctness pass-rate reward then a gated time-complexity reward (judge system compares generated solutions against reference optimal complexity); lightweight agentic RL for tool-use and search; deep-search training with turn-level judgment (logical soundness, tool-call accuracy, informational gain) and trajectory-level rewards, enabling stable long-horizon interaction of up to 600 tool-call turns[12]
Distillationnone disclosed

Modality

Typetext-only
Attachmentn/a[13]

Inference efficiency

KV-cache designnot disclosed
Quantisation shippednone disclosed
Speculative decodingnone disclosed
Serving optimisationsnone disclosed

Evaluation

Benchmarks (report's own numbers only)

BenchmarkValueRef
LiveCodeBench-V676.9Table 4 (Nanbeige4.1-3B)
LCB-Pro-Easy81.4Table 4 (Nanbeige4.1-3B)
LCB-Pro-Medium28.1Table 4 (Nanbeige4.1-3B)
AIME 2026 I87.4Table 4 (Nanbeige4.1-3B)
HMMT Nov77.92Table 4 (Nanbeige4.1-3B)
IMO-Answer-Bench53.38Table 4 (Nanbeige4.1-3B)
GPQA83.8Table 4 (Nanbeige4.1-3B)
HLE (Text-only)12.6Table 4 (Nanbeige4.1-3B)
Arena-Hard-V273.2Table 4 (Nanbeige4.1-3B)
Multi-Challenge52.21Table 4 (Nanbeige4.1-3B)
BFCL-V456.5Table 4 (Nanbeige4.1-3B)
Tau2-Bench48.57Table 4 (Nanbeige4.1-3B)
GAIA (text-only)69.9Table 5 (Nanbeige4.1-3B)
BrowseComp19.12Table 5 (Nanbeige4.1-3B)
BrowseComp-ZH31.83Table 5 (Nanbeige4.1-3B)
HLE (text-only)22.29Table 5 (Nanbeige4.1-3B)
SEAL-041.44Table 5 (Nanbeige4.1-3B)
xBench-DeepSearch-0575.0Table 5 (Nanbeige4.1-3B)
xBench-DeepSearch-1039.0Table 5 (Nanbeige4.1-3B)
LeetCode Weekly Contests 484-488 (pass rate)85.0Table 6 (Nanbeige4.1-3B)

Provenance references

  1. stated · Title ('Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts')
  2. stated · Title page byline
  3. stated · pipeline
  4. stated · Abstract ('The model checkpoint is available at https://huggingface.co/Nanbeige/Nanbeige4.1-3B'); §1 ('We open-source Nanbeige4.1-3B')
  5. stated · Abstract ('a unified generalist language model that ... with only 3B parameters')
  6. n/a · §2.1.1
  7. stated · §2.1.1
  8. stated · §2.1.1, §2.2.1, §2.3.2
  9. stated · §2.4, §2.3.2
  10. stated · §2.1.1, §2.4
  11. stated · §2.1.2
  12. stated · §2.1.2, §2.1.3, §2.2.2, §2.3.3, §2.4
  13. n/a · §1